runPredictions: Fit SGL Model, Predict Responses, and Assess Prediction...

In kouroshz/creNet: Fit a GLM with a combination of lasso and overlapping group lasso regularization

Usage

runPredictions(ents, rels, x.train, y.train, x.test = NULL, y.test = NULL, type = c("linear", "logit"), alpha = 0.95, nlam = 20, min.frac = 0.05, cv = TRUE, nfold = 10, cre = c("filter", "weight", "both", "none"), cre.sig = 0.01, standardize = c("all","self","train","no"), cores = 1, verbose = TRUE)

Arguments

ents	Entry data frame typically created by processMicroarray
rels	Relation data frame typically created by processMicroarray
x.train	Vector of responses for training data
y.train	Matrix of covariates for training data
x.test	Optional matrix of covariates for testing data
y.test	Optional vector of responses for testing data
type	Type of regression model: linear or logit
alpha	Tradeoff between lasso penalty and group lasso penalty. alpha=1 is pure lasso, alpha=0 is pure group lasso. Several values can be specified
nlam	Number of lambda values for the regularization path
min.frac	Smallest lambda value as a fraction of the largest
cv	logical flag: should the data be cross-validated?
nfold	Number of folds for cross-validation
cre	CRE method for filtering and/or computing group weights
cre.sig	significance level for CRE filtering
standardize	type of standardization
cores	Number of cores to be used in computations. cores>1 will result in parallel computations
verbose	logical flag for verbosity level

Details

The possible values of standardization are: "all": training and testing data are concatenated and then standardized, "self": each data set (training and testing) is standardized separately, "train": both training and testing data are standardized using the means and scale of the training data, "no": no standardization.

Value

A list with components

fit	Fitted object(s) of class creSGL if cv=FALSE or cvcreSGL if cv=TRUE; one fitted object per value of alpha
alpha	Input argument alpha
bestlam	Best value(s) of $lambda$ for cross-validation score, NULL otherwise; bestlam has same length as alpha
pred	Vector/matrix of predictions for training data and for testing data if specified; each column corresponds to a value of alpha
accuracy	Accuracy measures in prediction
slice	Duplicated matrix of covariates for training data and for testing data if available

Author(s)

Kourosh Zarringhalam and David Degras

kouroshz/creNet documentation built on May 20, 2019, 1:11 p.m.